The present invention relates to an image area discriminating device for a digital copier, facsimile device, scanner or similar imaging equipment and, more particularly, to an image area discriminating device capable of automatically determining whether or not areas constituting an input image each has undergone dot processing.
A digital copier, for example, reads a document image pixel by pixel by using a CCD (Charge Coupled Device) image sensor or line sensor. An analog electric signal appearing on the output of the image CCD image sensor is converted into a digital signal and then subjected to various kinds of processing. The processed signal is fed to a recording device to print out the document image on a recording medium. It is a common practice with such a digital copier to record the processed signal in two levels, i.e., record/non-record or in multiple levels since the recording device is unable to readily change the density level pixel by pixel. However, the copier has to reproduce even halftone images which are often carried on photographs and other similar documents. Conventional implementations for rendering halftone by use of the bilevel or multilevel recording device include a dither method, density pattern method, submatrix method, and error scattering method.
So long as the density of an image changes slowly as is the case with a photograph, conventional halftone processing successfully reproduces the image relatively attractively. However, when a character or similar image whose density changes in definite two levels is reproduced, the reproduction suffers from various defects such as illigibly blurred contours and contaminated background. For this reason, characters or similar document images should be subjected to simple bilevel or multilevel processing rather than halftone processing. If the copier is provided with an extra switch for entering the presence/absence of halftone on a document, the operator can select an adequate copy mode matching the document by operating the switch. In practice, however, many documents, typically pamphlets, carry both of halftome images and bilevel images such as characters. Then, selecting the bilevel or multilevel mode would lower quality of resultant photographs, and selecting the halftone mode would lower the quality of resultant characters.
Another and serious problem with this type of digital copier is that when the change in the density of a document has periodicity, the period or pitch of the density change and the pitch at which the cells of a CCD line sensor or similar image sensor are arranged, i.e., sampling period interfere with each other to cause moire to appear on a recorded image. For example, an image printed on a document in the form of dots has periodicity concerning the density change, so that the period of the density change interferes with the sampling frequency of the line sensor to bring about the moire problem. Specifically, assuming that the resolution of the line sensor is 400 dots per inch (dpi), moire is apt to appear in an image signal when dots are printed on a document in a density close to the resolution, i.e., 133 lines (about 10.5 pixels per millimeter)to 200 lines (about 16 pixels per millimeter). While densities outside the above-mentioned range also cause moire to occur, moire is noticeable in such a particular range and causes ::the signal to change over a broad range. Dot printing itself is a kind of quasi-halftone representation and changes the density in two levels, i.e., ONE and ZERO (recoRd and non-record) concerning each pixel. To render halftone, dot printing changes the average density of the entire assembly of pixels in multiple levels by changing the pitch or the size of dots. Therefore, apart from the moire problem, a dot image will be desirably reproduced if the signal is subjected to bilevel processing. In practice, however, a document on which dots are prined in a particular density suffers from moire when reproduced, as stated above.
On the other hand, when an image signal representative of a document is transformed to a bilevel or multilevel signal halftone processing, moire does not appear or appears little on reproduction since the halftone processing includes a step of averaging the densities of a plurality of pixels and a step of changing the threshold level. In this case, although the densities of a reproduction are rendered in quasi-halftone by dots, the dots on the reproduction are generated by halftone processing particular to a copier and, therefore, not the faithful replica of the dots on the document. It follows that when the image to be reproduced is an assembly of printed dots or an image reproduced by dot processing by a digital copier, a copy mode which effects halftone processing is preferable despite that such an image is bilevel concerning pixels.
It is desirable to process character portions by the simple bilevel or multilevel scheme and to process dot portions by the dither scheme or similar halftone processing scheme, as described previously. For this purpose, a document may be divided on an area basis. Specifically, if dot areas are detected and subjected to halftone processing while the other areas are subjected to simple bilevel processing, characters and photographs which are rendered by dots will be reproduced attractively, as disclosed in Japanese Patent Laid-Open Publication No. 279665/1988, for example. The method disclosed in this Laid-Open Publication is such that a bidimensional pattern of input image data is compared with a predetermined pattern to detect record dots and non-record dots and, based on the result of detection, whether or not the input image data is representative of a dot pattern is determined.
Now, in an image undergone dot processing, record dots such as black pixels and non-record dots such as white pixels are arranged alternately at a predetermined pitch and a predetermined distance. Assume that a record pixel located at a given position and non-record pixels surrounding it are arranged in a particular pattern, or that a non-record pixel located at a given position and record pixels surrounding it are arranged in a particular pattern. If such a condition appears repetitively, the pixel of interest surrounded by the non-record pixels or the record pixels can be regarded as a pixel undergone dot processing. Therefore, whether or not an input image is a dot pattern can be determined if image data lying in a bidimensional area constitued by a pixel of interest, which is sequentially shifted, and surrounding pixels is compared with a predetermined record dot detection pattern and a non-record dot detection pattern. However, when an image undergone dot processing is actually read by an image scanner, the image pattern of the resultant signal noticeably changes to obstruct accurate identification of dots. This stems from the fact that since dot printing renders density in terms of the area of record dots in a predetemined small area, a change in the density of an image translates into a noticeable change in the shape of the dots. Especially, when the dot density is around 50% or so, nearby record dots such as black pixels or non-record dots such as white pixels are sometimes connected together. Then, neither the record dots nor the non-record dots can be detected.
To reduce the discrimination error particular to the dot density of 50%, the thresold level for binarizing image data into the record and non-record pixel levels may be adjusted. This, however, aggravates discrimination error when the dot density is higher or lower than 50%. To eliminate this dilemmatic situation, use may be made of at least two different threshold values, and two independent circuits for detecting record dots and non-record dots, respectively. Then, a dot pattern is identified on the basis of the results of detection of record dots and non-record dots with reference made to the image data derived from the two differnt threshold values.
Generally, when a dot image is read by a scanner, the resultant signal appears as shown in FIG. 14. As FIG. 14 indicates, the signal changes in the height of peaks and the depth of troughs as well as in duty in association with the density. Paying attention to the signal with a density level of 50%, for example, the height of a peak and the depth of a trough changes with the position of the image. When the signal with the density of 50% is binarized at the threshold level TH.sub.1, the leading portion Pa has a peak higher than the level TH.sub.1 an a trough lower than the level TH.sub.1 and, therefore, the peak and the trough appear as, respectively, a record pixel and a non-record pixel in the resultant bilevel signal. However, in the trailing portion Pb, the peak and the trough both are higher than the threshold level TH.sub.1 with the result that a non-record pixel does not appear in the resultant bilevel signal. More specifically, such a signal is binarized at the threshold level TH.sub.1, a dot (record dot) is detected in the leading portion Pa out of a record pixel and non-record pixel pattern, but no dots can be detected in the trailing portion Pb. Assume that the signal with the density of 50% is binarized at the other threshold level TH.sub.2. Then both of the peak and trough in the leading portion Pa are lower than the level TH.sub.2, so that a record dot does not appear in the bilevel signal; in the trailing portion, since the peak is higher than the level TH.sub.2 and the trough is lower than the level TH.sub.1, the peak and the trough appear as, respectively a record pixel and a non-record pixel in the bilevel signal. In this manner, the threshold TH.sub.2 allows a dot (non-record dot) to be detected in the trailing portion Pb out of a record pixel and non-record pixel pattern while failing to do so when it comes to the leading portion Pa.
It follows that even with a dot image whose density is 50% either a record dot or non-record dot can be detected if the threshold values TH.sub.1 and TH.sub.2 are used to detect a dot which is a record dot and to detect a dot which is a non-record dot, respectively. When the density is low such as 20%, record dots are detected by use of the threshold value TH.sub.1 while, when the density is high such as 80%, non-record dots are detected use of the threshold value TH.sub.2.
However, the conventional approach described above has some problems left unsolved, as follows. When record dots and non-record dots in a dot area do not appear as dots or appear in a defective pattern due to moire, they are not regarded dots. The conventional approach is extremely susceptible to noise and is apt to regard even noise due to a solitary point of low density as a dot, resulting in frequent detection error. When a dot document having the same pitch and size in the main and subscanning directions is read, the density amplitude (MTF) usually differs from the main scanning direction to the subscanning direction, although the difference depends on the reading method and characteristic of the system. Further, it is likely that an area containing, for example, a line extending in the 45.degree. direction is misidentified.
When the magnification is changed, the configuration of dot pattern is changed. Then, prepared patterns would fail to follow such a change in the configuration of a dot pattern and would thereby aggravate the misidentification of dots in the case of magnifications other than x1. Although this problem may be eliminated if the number of dot detection patterns is increased, this is not practicable without increasing the number of circuit components and elements.